I posted this in another thread before. A mathematician offering his services. Maybe it's something for you?

Could you develop a fast algorithm to calculate the discrete cosine transform (or any daubechies wavelet) of a non-power-of-two amount of samples?
I need it to be ridiculously fast; perfect accuracy is not crucial.
I'm working on a video codec, which is working extremely well so far. I get very competitive compression ratio's compared to H265, but the computational complexity is just ridiculous. (realtime is far from possible, except for very low resolution video)
I need to do a DCT (or DWT) of sample sizes typically between 64 and 1024, usually around 700. It's fine if you come up with different algorithms, depending on the sample size.

Pseudocode is fine; preferably C/C++.

I'm willing to pay you a decent sum, if it's fast enough.